软硬件环境

• ubuntu 18.04 64bit
• NVIDIA GTX 1070Ti 8G
• anaconda with python 3.6
• opencv 3.4.3
• cuda 9.0
• YOLO v3

前言

下图是近年来物体检测领域算法的演化，YOLO是目前公认的比较准确的物体检测算法，已经发展到了第三个版本。关于darknet(实现YOLO检测的开源项目)的基本情况，参考之前的博文 https://xugaoxiang.com/2019/12/16/darknet-basic/，里面有比较详细的阐述。

准备工作

下载YOLO检测需要用到的配置文件、weights模型文件及物体类型class文件

wget https://pjreddie.com/media/files/yolov3.weights
wget https://github.com/pjreddie/darknet/blob/master/cfg/yolov3.cfg?raw=true -O ./yolov3.cfg
wget https://github.com/pjreddie/darknet/blob/master/data/coco.names?raw=true -O ./coco.names

YOLO的基本原理

一般来讲，物体检测由两部分组成，物体定位(object locator)和物体识别(object recognizer)。下面以图片为例来讲YOLO的实现原理

1. 将图片分割成13x13大小的网格单元, 以一张416x416像素大小的图片为例，会有1024个网格单元，模型会在每个cell上预测bounding box
2. 每个cell有可能会被预测出多个bounding box,大部分bounding box最后都会被清除，因为它们的相似度太低。这里有个算法叫Non-Maximum Suppression，翻译过来叫非极大值抑制，可参考论文Efficient-Non-Maximum-Suppression,　NMS算法是用来提取相似度最高的

python代码

opencv 3.4.2及以上的版本已经支持darknet，同时也支持使用其他常见深度学习框架的模型，如torch、tensorflow、caffe等。

# -*- coding: utf-8 -*-
# @time    : 18-10-26 下午4:47
# @author  : xugaoxiang
# @email   : xugx.ai@gmail.com
# @website : https://xugaoxiang.com
# @file    : opencv_yolov3.py
# @software: PyCharm

# Usage example: python3 opencv_yolov3.py --image=test.png

import sys
import cv2
import argparse
import numpy as np
import os.path

# 参数初始化
# 相似度阈值
confThreshold = 0.5  # Confidence threshold

# NMS算法阈值
nmsThreshold = 0.4

# 输入图片的宽和高
inpWidth = 416  
inpHeight = 416

parser = argparse.ArgumentParser(description = 'Object detection using YOLOv3 in opencv')
parser.add_argument('--image', help = 'Path to image file.')
args = parser.parse_args()

# 导入物体类别class文件，默认支持80种
classesFile = "coco.names"
classes = None
with open(classesFile, 'rt') as f :
    classes = f.read().rstrip('\n').split('\n')

# yolo v3的配置及weights文件
modelConfiguration = "yolov3.cfg"
modelWeights = "yolov3.weights"

# opencv读取外部模型
net = cv2.dnn.readNetFromDarknet(modelConfiguration, modelWeights)
net.setPreferableBackend(cv2.dnn.DNN_BACKEND_OPENCV)
# 这里使用CPU，如果想使用GPU的话，参数是DNN_TARGET_OPENCL,　但是当前版本只支持interl GPU，如果是其它GPU的话，会自动切换到CPU模式
net.setPreferableTarget(cv2.dnn.DNN_TARGET_CPU)

# Get the names of the output layers
def getOutputsNames(net) :
    # Get the names of all the layers in the network
    layersNames = net.getLayerNames()
    # Get the names of the output layers, i.e. the layers with unconnected outputs
    return [layersNames[i[0] - 1] for i in net.getUnconnectedOutLayers()]

# 画bounding box
def drawPred(classId, conf, left, top, right, bottom) :
    # Draw a bounding box.
    cv2.rectangle(frame, (left, top), (right, bottom), (255, 178, 50), 3)

    label = '%.2f' % conf

    # Get the label for the class name and its confidence
    if classes :
        assert (classId < len(classes))
        label = '%s:%s' % (classes[classId], label)

    # Display the label at the top of the bounding box
    labelSize, baseLine = cv2.getTextSize(label, cv2.FONT_HERSHEY_SIMPLEX, 0.5, 1)
    top = max(top, labelSize[1])
    cv2.rectangle(frame, (left, top - round(1.5 * labelSize[1])), (left + round(1.5 * labelSize[0]), top + baseLine),
                 (255, 255, 255), cv2.FILLED)
    cv2.putText(frame, label, (left, top), cv2.FONT_HERSHEY_SIMPLEX, 0.75, (0, 0, 0), 1)

# 使用NMS算法，丢弃低相似度的bounding box
def postprocess(frame, outs) :
    frameHeight = frame.shape[0]
    frameWidth = frame.shape[1]

    classIds = []
    confidences = []
    boxes = []
    # Scan through all the bounding boxes output from the network and keep only the
    # ones with high confidence scores. Assign the box's class label as the class with the highest score.
    classIds = []
    confidences = []
    boxes = []
    for out in outs :
        for detection in out :
            scores = detection[5 :]
            classId = np.argmax(scores)
            confidence = scores[classId]
            if confidence > confThreshold :
                center_x = int(detection[0] * frameWidth)
                center_y = int(detection[1] * frameHeight)
                width = int(detection[2] * frameWidth)
                height = int(detection[3] * frameHeight)
                left = int(center_x - width / 2)
                top = int(center_y - height / 2)
                classIds.append(classId)
                confidences.append(float(confidence))
                boxes.append([left, top, width, height])

    # Perform non maximum suppression to eliminate redundant overlapping boxes with
    # lower confidences.
    indices = cv2.dnn.NMSBoxes(boxes, confidences, confThreshold, nmsThreshold)
    for i in indices :
        i = i[0]
        box = boxes[i]
        left = box[0]
        top = box[1]
        width = box[2]
        height = box[3]
        drawPred(classIds[i], confidences[i], left, top, left + width, top + height)

# Process inputs
winName = 'Deep learning object detection in OpenCV'
cv2.namedWindow(winName, cv2.WINDOW_NORMAL)

if (args.image) :

    if not os.path.isfile(args.image) :
        print('Input image file {} does not exist.'.format(args.image))
        sys.exit(1)
    frame = cv2.imread(args.image, cv2.IMREAD_ANYCOLOR)
    outputFile = args.image[:-4] + '_yolov3_out.png'

    # Create a 4D blob from a frame.
    blob = cv2.dnn.blobFromImage(frame, 1 / 255, (inpWidth, inpHeight), [0, 0, 0], 1, crop = False)

    # Sets the input to the network
    net.setInput(blob)

    # Runs the forward pass to get output of the output layers
    outs = net.forward(getOutputsNames(net))

    # Remove the bounding boxes with low confidence
    postprocess(frame, outs)

    cv2.imshow(winName, frame)
    cv2.imwrite(outputFile, frame)
    cv2.destroyAllWindows()

测试程序输出

参考资料

• https://github.com/pjreddie/darknet/wiki/YOLO:-Real-Time-Object-Detection
• https://docs.opencv.org/3.4.3/da/d9d/tutorial_dnn_yolo.html
• https://xugaoxiang.com/2019/12/16/darknet-basic/